The entire post was enjoyable but I found the last paragraph to have the most actionable advice:
What’s much more useful is recording what the deep insights are, and storing them for recollection later. Because every important mathematical idea has a deep insight, and these insights are your best friends. They’re your mathematical “nose,” and they’ll help guide you through the mansion.
Spot on, about caring about conversions, not the click through rate.
Also true that this seems like a nuclear grade weapon. We have a system at optimine that is similar in a lot of ways, but also a lot less complex. Of course we are representing the advertiser side of the equation so, completely different design goals.
I wouldn't classify this as "nuclear grade", not in the least. I've seen the gentleman in the cubicle next to mine use more sophisticated simulations than this to model whether there will be fresh coffee in the pot when he gets to work.
In fact, I was surprised at the simplicity of the technique they demonstrate. Logistic regression is a very powerful method, but it is generally chosen because it is simple to implement and both fast to train and reliable to train (there are no issues with whether or when it will converge).
Of course, as is almost always true of machine learning, knowing how the mechanism works is almost completely unhelpful to duplicate the results. Although this technique is does some pretty innovative things, it would be pretty trivial to achieve the same quality results if you didn't know their technique, but you did know what the feature set was (including any preprocessing and cleanup done on features). However, this system seems designed to give good enough results with extremely agressive performance SLAs, which I suspect is very nontrivial indeed.
...and the counterargument has been debunked too. A lot of complaints were just ad hominem attack alleging Tamil supremacy, Tamil ethnocentrism. The first note of dicord that strikes you as an Indian is that bar one, all authors of that paper were from North India. If any Tamil bias is expected from that rergion it would be a bias against Tamil.
I dont know why Rao's claim (well they arent quite claims either, not yet atleast, they are rather a call for further investigation) have been so spectacularly blown out of proportion and why people get so upset about it.
Sproat at least does not say that Rao in any form claimed that his work "proves" anything one way or the other, rather that it was the "discussion" around the paper that claims a proof. I would have been happier if that distinction was made clearer.
In anycase if you search HN you will find an interesting thread discussing this topic. Learned quite a bit from it.
EDIT: @kylebgorman I dont consider myself qualified enough to agree or disagree, but have to say that I was taken aback by the push back it received, particulay the vociferous allegation of Tamil supremacy.
EDIT @kylebgorman wait I didnt say that the paper or the criticism was ethnically biased, but that ethnic bias was a major criticism that was levied against Rao's paper. This comment on the thread will have some examples http://news.ycombinator.com/item?id=4062129 rebuttals, counter rebuttals, , counter-counter... you get the idea.
The many computational linguists who have discussed these papers in public fora have expressed disgust with the scientific naïveity of the Rao et al. paper; the pushback is due to its very poor scientific merits (in contrast with its very high publication profile), not some ethnic bias as you seem to allege.
Decision trees have proven to be sucessful in direct marketing/ecommerce applications such as incremental response modeling. Because of their explicit nature, decision trees trees can be used for workflow optimization in industrial environments as well.
I'm hoping to write a couple of follow-up posts talking about some of the trade-offs that you're going to have to make when you're a startup that doesn't have the resources of a Facebook or a LinkedIn. It's difficult to do everything in that (admittedly great) article if your resources are limited.
Thanks, I didn't see that. I tend to like the back button, I just think they could have done it better. I think I probably will end up putting together some type of array driven navigation, where I can provide the workflow ahead of time.
Very cool. It will be interesting to see how successful AppHarbor can be without first establishing a competitor to Azure Appfabric.The service bus and access control inherit in Appfabric makes cloud computing palatable for enterprise companies.